METHODS

We used the FEEDcities 2021 mapping of SF vendors (n=371) as the sampling frame for the present study.9 From that frame, 120 vendor sites were randomly selected to serve as the vendor list for 2024 fieldwork. To ensure probability proportional to size (PPS) sampling across the city, 100 geographic sampling areas were defined based on district population size, and selected vendors were assigned to these areas. At each assigned vendor, fieldworkers recruited approximately ten adult consumers using convenience sampling. Eligibility criteria included adults aged 18 years or older who had resided in Tbilisi for at least one year. Data were collected through structured, face-to-face interviews. Verbal informed consent was obtained from all participants. Of 900 individuals approached, 860 provided complete and valid data after cleaning.

Study variables

The outcome variable was overweight or obesity, defined as a Body Mass Index (BMI) > 25 kg/m², calculated from self-reported weight and height. WHO classification standards were applied, categorizing individuals as normal weight (18.5-24.9 kg/m²), overweight (25.0-29.9 kg/m²), and obese (≥30.0 kg/m²).10 Because underweight cases were rare, these individuals were grouped with the normal-weight category for analysis. For regression models, BMI was dichotomized into <25 kg/m² (normal weight) and≥25 kg/m² (overweight or obese). The key exposure variables included SF consumption behaviors, lifestyle factors, health-related perceptions, sociodemographic characteristics, and health status.

​

Both frequency and diversity assessed SF consumption. Frequency was based on self-reported weekly consumption. Participants were categorized as low-frequency consumers (<2 times per week) or high-frequency consumers (≥2 times per week). Diversity of SF consumption was measured as the number of distinct food categories consumed in the past month. For this purpose, SF were classified into five major groups based on predominant ingredients and preparation methods. These categories were developed in the absence of a standardized national food composition database for SF in Georgia and were therefore informed by culinary characteristics. The groups included: (a) Khachapuri, traditional Georgian cheese-filled bread, represented by four regional variations (Megruli, Imeruli, Guruli, and layered khachapuri); (b) Lobiani, Georgian bean-filled bread, available in two types (traditional lobiani and lobiani with ham); (c) Sweet pastries, such as baklava, cream cake, loose cake or muffin, sweet puff pastry, and sweet buns; (d) Savory baked goods, including kubdari (a meat-filled bread), savory meat pie, savory potato pie, and chebureki (a deep-fried pastry filled with meat); and (e) Non-traditional fast foods, such as hamburgers, hotdogs, shawarma, and pizza. Participants consuming items from zero to two of these categories were classified as low-diversity consumers, whereas those consuming from three to five categories were considered high-diversity consumers.

​

Sugar-sweetened beverage (SSB) intake was analyzed as part of consumption behaviors. Participants were classified as rare consumers (<1 time per week), occasional consumers (1-3 times per week), or regular consumers (≥4 times per week).

Physical activity and sedentary behavior were self-reported as minutes per day and analyzed as continuous variables (minutes/day). Physical activity was represented by the minutes spent engaging in moderate-to-vigorous activity (e.g., walking, running, cycling). In contrast, sedentary behavior was represented by the minutes spent sitting or lying awake (excluding sleep).

​

Perceived health risks of SF were measured using five 5-point Likert items addressing perceived fat quantity, fat quality, reheated-oil use, disease risk, and weight gain. Items were coded from 1 to 5 and reverse-scored where appropriate so that higher values indicated greater perceived risk. Item scores were summed to produce a total perception score (range 5-25). For descriptive analysis, scores were grouped into three categories: low (<15), moderate (15-19), and high (≥20). For multivariable models, the score was entered as a continuous covariate. The internal consistency of the scale was evaluated using Cronbach’s α (α ≥ 0.70 considered acceptable).

​

Health status variables included self-reported diagnoses of chronic conditions. A composite “any chronic condition” indicator was created (Yes/No), coded “Yes” if participants reported at least one of the following: hypertension, high cholesterol, diabetes, heart disease, or cancer (self-reported diagnoses). Sociodemographic characteristics were assessed using indicators of education, income, and employment. To strengthen analyses, a composite socioeconomic status (SES) variable was created by combining education and income. Education was categorized as low (no formal schooling, refused, or primary), middle (secondary or vocational), and high (college or university). Income was grouped as low (<500 GEL or none), middle (500-1,499 GEL), and high (≥1,500 GEL). Each variable was coded 1-3 and summed (range 2-6) to construct an SES score. SES was then classified as low (2-3), mid (4), and high (5-6). Participants missing either education or income were coded as missing on SES.

Data analysis

Descriptive statistics were computed for all variables. Means and standard deviations (SD) were reported for continuous variables, while frequencies and percentages were presented for categorical variables. To identify factors associated with overweight and obesity, several statistical tests were applied. Associations between categorical predictors and overweight/obesity (BMI ≥25 vs. <25 kg/m²) were evaluated using chi-square tests. For continuous predictors (age, physical activity, and sedentary time), mean differences by BMI category were assessed using independent-samples t-tests.

​

To determine variables independently associated with overweight/obesity, binary logistic regression analysis was conducted. Regression coefficients (B) represented the change in the log-odds of being overweight/obese for each predictor. At the same time, odds ratios (Exp(B)) indicated how the odds of the outcome changed with a one-unit increase in the predictor variable. Statistical significance was defined as p<0.05. Analyses were conducted in SPSS Version 16.0 and Microsoft Excel 2016.

RESULTS